Merge pull request #1930 from martin-frbg/issue1908

Reflect ARMV8 target definition changes from PR1876
This commit is contained in:
Martin Kroeker 2018-12-23 15:06:33 +01:00 committed by GitHub
commit 942e229ed5
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 111 additions and 5 deletions

View File

@ -116,18 +116,19 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
"#define L2_LINESIZE\t64\n"
"#define DTB_DEFAULT_ENTRIES\t64\n"
"#define DTB_SIZE\t4096\n"
"#define L2_ASSOCIATIVE\t32\n")
"#define L2_ASSOCIATIVE\t32\n"
"#define ARMV8\n")
set(SGEMM_UNROLL_M 4)
set(SGEMM_UNROLL_N 4)
elseif ("${CORE}" STREQUAL "CORTEXA57")
elseif ("${CORE}" STREQUAL "CORTEXA57" OR "${CORE}" STREQUAL "CORTEXA53")
file(APPEND ${TARGET_CONF_TEMP}
"#define L1_CODE_SIZE\t49152\n"
"#define L1_CODE_SIZE\t32768\n"
"#define L1_CODE_LINESIZE\t64\n"
"#define L1_CODE_ASSOCIATIVE\t3\n"
"#define L1_DATA_SIZE\t32768\n"
"#define L1_DATA_LINESIZE\t64\n"
"#define L1_DATA_ASSOCIATIVE\t2\n"
"#define L2_SIZE\t2097152\n"
"#define L2_SIZE\t262144\n"
"#define L2_LINESIZE\t64\n"
"#define L2_ASSOCIATIVE\t16\n"
"#define DTB_DEFAULT_ENTRIES\t64\n"
@ -135,7 +136,8 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
"#define HAVE_VFPV4\n"
"#define HAVE_VFPV3\n"
"#define HAVE_VFP\n"
"#define HAVE_NEON\n")
"#define HAVE_NEON\n"
"#define ARMV8\n")
set(SGEMM_UNROLL_M 16)
set(SGEMM_UNROLL_N 4)
set(DGEMM_UNROLL_M 8)
@ -144,6 +146,109 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
set(CGEMM_UNROLL_N 4)
set(ZGEMM_UNROLL_M 8)
set(ZGEMM_UNROLL_N 4)
elseif ("${CORE}" STREQUAL "CORTEXA72" OR "${CORE}" STREQUAL "CORTEXA73")
file(APPEND ${TARGET_CONF_TEMP}
"#define L1_CODE_SIZE\t49152\n"
"#define L1_CODE_LINESIZE\t64\n"
"#define L1_CODE_ASSOCIATIVE\t3\n"
"#define L1_DATA_SIZE\t32768\n"
"#define L1_DATA_LINESIZE\t64\n"
"#define L1_DATA_ASSOCIATIVE\t2\n"
"#define L2_SIZE\t524288\n"
"#define L2_LINESIZE\t64\n"
"#define L2_ASSOCIATIVE\t16\n"
"#define DTB_DEFAULT_ENTRIES\t64\n"
"#define DTB_SIZE\t4096\n"
"#define HAVE_VFPV4\n"
"#define HAVE_VFPV3\n"
"#define HAVE_VFP\n"
"#define HAVE_NEON\n"
"#define ARMV8\n")
set(SGEMM_UNROLL_M 16)
set(SGEMM_UNROLL_N 4)
set(DGEMM_UNROLL_M 8)
set(DGEMM_UNROLL_N 4)
set(CGEMM_UNROLL_M 8)
set(CGEMM_UNROLL_N 4)
set(ZGEMM_UNROLL_M 8)
set(ZGEMM_UNROLL_N 4)
elseif ("${CORE}" STREQUAL "FALKOR")
file(APPEND ${TARGET_CONF_TEMP}
"#define L1_CODE_SIZE\t65536\n"
"#define L1_CODE_LINESIZE\t64\n"
"#define L1_CODE_ASSOCIATIVE\t3\n"
"#define L1_DATA_SIZE\t32768\n"
"#define L1_DATA_LINESIZE\t128\n"
"#define L1_DATA_ASSOCIATIVE\t2\n"
"#define L2_SIZE\t524288\n"
"#define L2_LINESIZE\t64\n"
"#define L2_ASSOCIATIVE\t16\n"
"#define DTB_DEFAULT_ENTRIES\t64\n"
"#define DTB_SIZE\t4096\n"
"#define HAVE_VFPV4\n"
"#define HAVE_VFPV3\n"
"#define HAVE_VFP\n"
"#define HAVE_NEON\n"
"#define ARMV8\n")
set(SGEMM_UNROLL_M 16)
set(SGEMM_UNROLL_N 4)
set(DGEMM_UNROLL_M 8)
set(DGEMM_UNROLL_N 4)
set(CGEMM_UNROLL_M 8)
set(CGEMM_UNROLL_N 4)
set(ZGEMM_UNROLL_M 8)
set(ZGEMM_UNROLL_N 4)
elseif ("${CORE}" STREQUAL "THUNDERX)
file(APPEND ${TARGET_CONF_TEMP}
"#define L1_CODE_SIZE\t32768\n"
"#define L1_CODE_LINESIZE\t64\n"
"#define L1_CODE_ASSOCIATIVE\t3\n"
"#define L1_DATA_SIZE\t32768\n"
"#define L1_DATA_LINESIZE\t128\n"
"#define L1_DATA_ASSOCIATIVE\t2\n"
"#define L2_SIZE\t167772164\n"
"#define L2_LINESIZE\t128\n"
"#define L2_ASSOCIATIVE\t16\n"
"#define DTB_DEFAULT_ENTRIES\t64\n"
"#define DTB_SIZE\t4096\n"
"#define HAVE_VFPV4\n"
"#define HAVE_VFPV3\n"
"#define HAVE_VFP\n"
"#define HAVE_NEON\n"
"#define ARMV8\n")
set(SGEMM_UNROLL_M 4)
set(SGEMM_UNROLL_N 4)
set(DGEMM_UNROLL_M 2)
set(DGEMM_UNROLL_N 2)
set(CGEMM_UNROLL_M 2)
set(CGEMM_UNROLL_N 2)
set(ZGEMM_UNROLL_M 2)
set(ZGEMM_UNROLL_N 2)
elseif ("${CORE}" STREQUAL "THUNDERX2T99)
file(APPEND ${TARGET_CONF_TEMP}
"#define L1_CODE_SIZE\t32768\n"
"#define L1_CODE_LINESIZE\t64\n"
"#define L1_CODE_ASSOCIATIVE\t8\n"
"#define L1_DATA_SIZE\t32768\n"
"#define L1_DATA_LINESIZE\t64\n"
"#define L1_DATA_ASSOCIATIVE\t8\n"
"#define L2_SIZE\t262144\n"
"#define L2_LINESIZE\t64\n"
"#define L2_ASSOCIATIVE\t8\n"
"#define L3_SIZE\t33554432\n"
"#define L3_LINESIZE\t64\n"
"#define L3_ASSOCIATIVE\t32\n"
"#define DTB_DEFAULT_ENTRIES\t64\n"
"#define DTB_SIZE\t4096\n"
"#define VULCAN\n")
set(SGEMM_UNROLL_M 16)
set(SGEMM_UNROLL_N 4)
set(DGEMM_UNROLL_M 8)
set(DGEMM_UNROLL_N 4)
set(CGEMM_UNROLL_M 8)
set(CGEMM_UNROLL_N 4)
set(ZGEMM_UNROLL_M 4)
set(ZGEMM_UNROLL_N 4)
endif()
# Or should this actually be NUM_CORES?
@ -163,6 +268,7 @@ if (DEFINED CORE AND CMAKE_CROSSCOMPILING AND NOT (${HOST_OS} STREQUAL "WINDOWSS
file(APPEND ${TARGET_CONF_TEMP}
"#define GEMM_MULTITHREAD_THRESHOLD\t${GEMM_MULTITHREAD_THRESHOLD}\n")
# Move to where gen_config_h would place it
file(MAKE_DIRECTORY ${TARGET_CONF_DIR})
file(RENAME ${TARGET_CONF_TEMP} "${TARGET_CONF_DIR}/${TARGET_CONF}")
else(NOT CMAKE_CROSSCOMPILING)